Autoencoder-based information content preserving data anonymization method and system

摘要

A method of providing an auto-encoder for anonymizing data associated with a population of entities is disclosed. The method includes providing a computer system with a memory storing specific computer-executable instructions for a neural network. The neural network includes an input layer of nodes; three or more layers of nodes; and an output layer of nodes to provide an encoded output vector. The second layer of nodes has more nodes than the first and third layers of nodes. The method also includes identifying a plurality of characteristics associated with the entities and preparing a plurality of input vectors that include a characteristic. The characteristics appear in the input vector as transformed numeric information from human recognizable text. The method includes training the neural network during a plurality of training cycles comprising: processing an input vector with the neural network to provide an encoded output vector; determining an output vector reconstruction error by calculating a function of the encoded output vector and the input vector; back-propagating the output vector reconstruction error back through the neural network; and recalibrating a weight to minimize the output vector reconstruction error. Additional neural networks are also disclosed. The outputs of the additional neural networks may be combined. Encoded output vectors may be compared to identify a common characteristic between two or more entities or to identify two or more entities with the common characteristic. An auto-encoder system for anonymizing data is also disclosed.